Alluvial Diagrams

class: center, middle, inverse, title-slide

# Alluvial Diagrams
### Ludmila Janda
### 2021

---

.tiny .remark-code { /*Change made here*/
  font-size: 50% !important;
}
</style>

# What are Alluvial Diagrams?
<img src="data:image/png;base64,#images/alluvial-fans.jpeg" width="600" />

---

# What are Alluvial Diagrams?
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-3-1.png" width="864" />

---

# When to Use Alluvial Diagrams

- When showing groups moving from one state of being to another

- When you have a reasonable number of groups and states of being

---

# Terminology

- **Axes** the different states that the graph shows movement between

---

# Terminology

- **Stratum** the groupings at each state/axis

---

# Terminology

- **Flow** the movement from one state/axis to another

- **Alluvium** the movement across all states/axes (all flows together)

- **Lode** intersection of one alluvium and one stratum

---

# Terminology

- **Axes** the different states that the graph shows movement between

- **Stratum** the groupings at each state/axis

- **Flow** the movement from one state/axis to another

- **Alluvium** the movement across all states/axes (all flows together)

- **Lode** intersection of one alluvium and one stratum

---

# Raw Data

```r
sim_data_pre_post <- read_csv(here::here("sim_data_pre_post.csv")) %>% 
  select(student, unit_title, assessment, score_level)

head(sim_data_pre_post, 5)
```

```
## # A tibble: 5 x 4
##   student unit_title                      assessment score_level
##     <dbl> <chr>                           <chr>            <dbl>
## 1       1 Unicorn Traits and Reproduction pre                  1
## 2       2 Unicorn Traits and Reproduction pre                  1
## 3       3 Unicorn Traits and Reproduction pre                  2
## 4       4 Unicorn Traits and Reproduction pre                  1
## 5       5 Unicorn Traits and Reproduction pre                  1
```

---

# Data Cleaning

- Reshape data to wide by axes

```r
sim_data_pre_post <- read_csv(here::here("sim_data_pre_post.csv")) %>% 
  dplyr::select(student, unit_title, assessment, score_level) %>% 
  dplyr::mutate(score_level = factor(score_level), 
         assessment = factor(assessment, levels = c("pre", "post"))) %>% 
  tidyr::pivot_wider(names_from = assessment, values_from = score_level)

head(sim_data_pre_post, 5)
```

```
## # A tibble: 5 x 4
##   student unit_title                      pre   post 
##     <dbl> <chr>                           <fct> <fct>
## 1       1 Unicorn Traits and Reproduction 1     4    
## 2       2 Unicorn Traits and Reproduction 1     3    
## 3       3 Unicorn Traits and Reproduction 2     3    
## 4       4 Unicorn Traits and Reproduction 1     4    
## 5       5 Unicorn Traits and Reproduction 1     4
```

---

# Using `to_lodes_form()`

```r
sim_data_pre_post <- 
  sim_data_pre_post %>% 
  ggalluvial::to_lodes_form(key = "assessment", axes = 3:4)

head(sim_data_pre_post, 5)
```

```
## # A tibble: 5 x 5
##   student unit_title                      alluvium assessment stratum
##     <dbl> <chr>                              <int> <fct>      <fct>  
## 1       1 Unicorn Traits and Reproduction        1 pre        1      
## 2       2 Unicorn Traits and Reproduction        2 pre        1      
## 3       3 Unicorn Traits and Reproduction        3 pre        2      
## 4       4 Unicorn Traits and Reproduction        4 pre        1      
## 5       5 Unicorn Traits and Reproduction        5 pre        1
```

---

# Plot Elements

`ggalluvial::geom_flow(alpha = 0.5)` - gives you flows from one axis to the next (useful to make somewhat transparent with alpha)

`geom_alluvium(aes(fill = final_outcome)` - gives you the alluvia across all the axes (useful to pick how to set the fill)

`ggalluvial::geom_stratum()` - gives you the strata

---

# Plot Code - Add Strata

```r
sim_data_pre_post %>% 
  ggplot(aes(x = assessment,         # categorical axis var (pre or post)
             stratum = stratum,      # categorical group var (score level)
             alluvium = alluvium)) + # individual/unit (student)
  ggalluvial::geom_stratum() 
```

---

# Plot Code - Add Strata

---

# Plot Code - Add Flows

```r
sim_data_pre_post %>% 
  ggplot(aes(x = assessment,             
             stratum = stratum,          
             alluvium = alluvium)) +     
  ggalluvial::geom_stratum() +
  ggalluvial::geom_flow(alpha = 0.5) 
```

---

# Plot Code - Add Flows

---

# Plot Code - Add fill

```r
sim_data_pre_post %>% 
  ggplot(aes(x = assessment,              
             stratum = stratum,  
             alluvium = alluvium,         
             fill = stratum)) +  
  ggalluvial::geom_stratum() + 
  ggalluvial::geom_flow(alpha = 0.5) 
```

---

# Plot Code - Add fill

---

# Plot Code - Reverse Stratum

```r
sim_data_pre_post %>% 
  ggplot(aes(x = assessment,              
             stratum = fct_rev(stratum),  
             alluvium = alluvium,         
             fill = fct_rev(stratum))) +  
  ggalluvial::geom_stratum() + 
  ggalluvial::geom_flow(alpha = 0.5) 
```

---

# Plot Code - Reverse Stratum

---

# Plot - Facet Wrap

```r
library(ggalluvial)

---

# Plot - Facet Wrap
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-20-1.png" width="864" />

---

# Plot - Add colors

```r
library(ggalluvial)

---

# Plot - Add colors 
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-22-1.png" width="864" />

---

# Plot - Beautify

```r
library(ggalluvial)

sim_data_pre_post %>% 
  ggplot(aes(x = assessment, 
             stratum = fct_rev(stratum), 
             alluvium = alluvium,
             fill = fct_rev(stratum), 
             label = fct_rev(stratum))) +
  ggalluvial::geom_stratum() + 
  ggalluvial::geom_flow(alpha = 0.5) + 
  facet_wrap(~unit_title, scales = "free_y") + 
  scale_fill_manual("Score Level",
                    values = c("1" = "#BFBFBF", "2" = "#55A5CC", 
                               "3" = "#30779C", "4" = "#004C6D")) +
  scale_y_continuous(labels = comma) +
  labs(x = "") +
  theme_minimal() + 
  theme(panel.border = element_blank(), 
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank(), 
        axis.line = element_line(colour = "grey"), 
        axis.text.x = element_text(size = 18), 
        axis.text.y = element_text(size = 14), 
        strip.text.x = element_text(size = 16), 
        legend.position = "bottom", 
        legend.title = element_text(size = 16),
        legend.text = element_text(size = 16))
```

---

# Plot - Beautify
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-24-1.png" width="864" />

---

# Plot with `geom_alluvium()` 
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/ggalluvial2-1.png" width="1008" />

.tiny[
Source: https://mdneuzerling.com/post/my-data-science-job-hunt/
]

---

# Raw Data

```r
job_outcomes <- read_csv(here::here("job_outcomes.csv"))

head(job_outcomes, 10)
```

```
## # A tibble: 10 x 5
##    job_id contact   `1st stage` `2nd stage` outcome 
##     <dbl> <chr>     <chr>       <chr>       <chr>   
##  1      1 job site  ghosted     <NA>        <NA>    
##  2      2 job site  ghosted     <NA>        <NA>    
##  3      3 recruiter coffee      no role     <NA>    
##  4      4 LinkedIn  ghosted     <NA>        <NA>    
##  5      5 network   coffee      ghosted     <NA>    
##  6      6 network   phone call  interview   withdrew
##  7      7 LinkedIn  phone call  withdrew    <NA>    
##  8      8 LinkedIn  ghosted     <NA>        <NA>    
##  9      9 internal  coffee      interview   rejected
## 10     10 LinkedIn  phone call  interview   rejected
```

---

# Data Cleaning

```r
job_outcomes <- read_csv(here::here("job_outcomes.csv"))

job_outcomes %>% 
  mutate(final_outcome = coalesce(outcome, `2nd stage`, `1st stage`)) %>% 
  head(10)
```

```
## # A tibble: 10 x 6
##    job_id contact   `1st stage` `2nd stage` outcome  final_outcome
##     <dbl> <chr>     <chr>       <chr>       <chr>    <chr>        
##  1      1 job site  ghosted     <NA>        <NA>     ghosted      
##  2      2 job site  ghosted     <NA>        <NA>     ghosted      
##  3      3 recruiter coffee      no role     <NA>     no role      
##  4      4 LinkedIn  ghosted     <NA>        <NA>     ghosted      
##  5      5 network   coffee      ghosted     <NA>     ghosted      
##  6      6 network   phone call  interview   withdrew withdrew     
##  7      7 LinkedIn  phone call  withdrew    <NA>     withdrew     
##  8      8 LinkedIn  ghosted     <NA>        <NA>     ghosted      
##  9      9 internal  coffee      interview   rejected rejected     
## 10     10 LinkedIn  phone call  interview   rejected rejected
```

---

# Plot with `geom_alluvium()`

```r
job_outcomes <- read_csv(here::here("job_outcomes.csv"))

job_outcomes %>% 
  mutate(final_outcome = coalesce(outcome, `2nd stage`, `1st stage`)) %>% 
  ggalluvial::to_lodes_form(key = "contact", axes = 2:5) %>% 
  ggplot(aes(x = contact, stratum = stratum, 
             alluvium = alluvium, label = stratum)) + 
  ggalluvial::geom_alluvium(aes(fill = final_outcome), 
                            color = "darkgrey", na.rm = TRUE) +
  ggalluvial::geom_stratum(na.rm = TRUE) +
  geom_text(stat = "stratum", na.rm = TRUE, size = 5) + 
  theme_minimal() + 
  theme(text = element_text(size = 20), 
        legend.position = "bottom") +
  labs(x = "", fill = "Final Outcome", 
       caption = "David Neuzerling @mdneuzerling") + 
  scale_fill_manual(values = c("ghosted" = "#F0E442", 
                               "no role" = "#CC79A7",
                               "withdrew" = "#0072B2", 
                               "rejected" = "#D55E00",
                               "offer" = "#009E73")) 
```

---

# Plot with `geom_alluvium()`
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-28-1.png" width="1008" />

---

# Contrast with `geom_flow()`
<img src="data:image/png;base64,#alluvial_slides_files/figure-html/unnamed-chunk-29-1.png" width="1008" />

---

# Examples

.tiny[
Source: https://www.nytimes.com/interactive/2021/06/22/us/elections/results-nyc-mayor-primary.html
]

---

# Examples

.tiny[
Source: http://www.economist.com/blogs/graphicdetail/2015/05/daily-chart-1?fsrc=scn/tw/te/bl/ed/seeking_safety&%3Ffsrc%3Dscn/=tw/dc
]

---

# Examples

- Better as a mosaic?

<img src="data:image/png;base64,#images/ex-titanic.png" width="600" />
.tiny[
Source: https://cran.r-project.org/web/packages/ggalluvial/vignettes/ggalluvial.html
]

---

# Examples

- Including gender in this graph gives us very little information

.tiny[
Source: https://digitalsplashmedia.com/2014/06/visualizing-categorical-data-as-flows-with-alluvial-diagrams/
]

---

# Examples

- Way too much going on here!

.tiny[
Source: https://digitalsplashmedia.com/2014/06/visualizing-categorical-data-as-flows-with-alluvial-diagrams/
]

---

# In Summary

- Ensure that your data fits the alluvial specifications

- Reshape the data to wide by axes if needed for `to_lodes_form()`

- Consider whether you want to highlight flows or alluvium

- Pay attention to your use of color (both in terms of variable type and whether your are coloring from starting state or ending state)

- HAVE FUN!

---

# Resources

ggalluvial CRAN vignette:
  https://cran.r-project.org/web/packages/ggalluvial/vignettes/ggalluvial.html

defining and taxonomizing alluvial diagrams:
  https://corybrunson.github.io/2019/09/13/flow-taxonomy/
  
Alternative way to make alluvial charts: 
  https://ggforce.data-imaginist.com/reference/geom_parallel_sets.html